|Publication number||US7058687 B2|
|Application number||US 09/863,941|
|Publication date||Jun 6, 2006|
|Filing date||May 22, 2001|
|Priority date||Apr 3, 2001|
|Also published as||US20030028580|
|Publication number||09863941, 863941, US 7058687 B2, US 7058687B2, US-B2-7058687, US7058687 B2, US7058687B2|
|Original Assignee||Sendmail, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (8), Referenced by (30), Classifications (11), Legal Events (11)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The present application is related to and claims the benefit of priority of the following commonly-owned provisional application(s): application Ser. No. 60/281,313, filed Apr. 3, 2001, entitled “E-Mail System with Methodology for Accelerating Mass Mailings”, of which the present application is a non-provisional application thereof. The present application is related to the following commonly-owned application(s): application Ser. No. 09/499,502, filed Feb. 7, 2000, entitled “Electronic Mail System with Improved Methodology for Processing Messages with Mailing Lists”. The disclosures of each of the foregoing applications are hereby incorporated by reference in their entirety, including any appendices or attachments thereof, for all purposes.
A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
1. Field of the Invention
The present invention relates generally to electronic mail (e-mail) systems and, more particularly, to improved methodology for processing automated e-mail messages sent to numerous recipients.
2. Description of the Background Art
Today, electronic mail or “e-mail” is a pervasive, if not the most predominant, form of electronic communication.
A typical e-mail delivery process is as follows. In the following scenario, Larry sends e-mail to Martha at her e-mail address: firstname.lastname@example.org. Martha's Internet Service Provider (ISP) uses an MTA, such as provided by Sendmail® for NT, available from Sendmail, Inc. of Emeryville, Calif. (With a lower case “s,” “sendmail” refers to Sendmail's MTA, which is one component of the Sendmail® Switch product line.)
The MTA, which is responsible for queuing up messages and arranging for their distribution, is the workhorse component of electronic mail systems. The MTA “listens” for incoming e-mail messages on the SMTP port, which is generally port 25. When an e-mail message is detected, it handles the message according to configuration settings, that is, the settings chosen by the system administrator, in accordance with relevant standards such as Request For Comment documents (RFCs). Typically, the mail server or MTA must temporarily store incoming and outgoing messages in a queue, the “mail queue.” Actual queue size is highly dependent on one's system resources and daily volumes.
MTAs, such as the commercially-available Sendmail® MTA, perform three key mail transport functions:
The overall process may be summarized as follows. E-mail is routed via SMTP servers, the so-called “Mail Transfer Agents” (MTA). Users write, send, and read e-mail via Mail User Agents (MUA). To send e-mail, an MUA connects to an MTA which receives the e-mail and routes it to another MTA. An intermediary MTA might forward the e-mail to yet another MTA until the e-mail reaches the destination system, where the e-mail is stored in a mailbox accessible by the recipient.
For further description of e-mail systems, see e.g., Sendmail® for NT User Guide, Part Number DOC-SMN-300-WNT-MAN-0999, available from Sendmail, Inc. of Emeryville, Calif., the disclosure of which is hereby incorporated by reference. Further description of the basic architecture and operation of e-mail systems is available in the technical and trade literature; see e.g., the following RFC (Request For Comments) documents:
RFC821 Simple Mail Transfer Protocol (SMTP) RFC822 Standard for the Format of ARPA Internet Text Messages RFC974 Mail Routing and the Domain System RFC937, RFC1081 Post Office Protocol version 3 (POP3) RFC1123 Requirements for Internet Hosts-Application and Support RFC1725 Post Office Protocol version 3 (POP3) RFC2033 Local Mail Transfer Protocol (LMTP) RFC2060, RFC2061 Internet Message Access Protocol (IMAP) RFC2246 The TLS Protocol, version 1.0 RFC2487 SMTP Service Extension for Secure SMTP over TLS
RFCs are numbered Internet informational documents and standards widely followed by commercial software and freeware in the Internet and UNIX communities. The RFCs are unusual in that they are floated by technical experts acting on their own initiative and reviewed by the Internet at large, rather than formally promulgated through an institution such as ANSI. For this reason, they remain known as RFCs even once they are adopted as standards. The above-listed RFC documents are currently available via the Internet (e.g., at http://www.ietf.org/rfc), the disclosures of which are hereby incorporated by reference.
Often when sending e-mail, a distribution or “mailing list” is employed to facilitate the process of sending an e-mail message to a group of people. For instance, instead of addressing an e-mail message to individual members of a recurring group, a user can instead simply define a mailing list to comprise those members. For example, the user could define a “Marketing” mailing list that specifies members of the marketing department of the user's company. Once defined, the mailing list can be used in the recipient field for an e-mail message, in lieu of listing individual members. A message sent to this distribution list goes to all recipients listed. Typically, e-mail systems provide graphical user interface facilities for managing (e.g., adding and deleting) names in a mailing list.
Expectedly, as a particular list grows larger, it becomes progressively more resource intensive and time consuming to manage and process. Although the foregoing example of a mailing list for a marketing department may comprise a comparatively small group of recipients (e.g., less than 100), a mailing list can in fact specify an extremely large group of recipients. Consider, for instance, a mailing list defined for customer support (e.g., “North American Users”) for a large software company. As another example, ISPs (Internet Service Providers) typically support many domains, many lists within each domain, and many users for each list. In such a case, a given mailing list may in fact specify many thousands or even millions of recipients, leading to an incredible amount of mailing list traffic. Accordingly, there is great interest in improving the management and processing of mailing lists so that e-mail sent to mailing lists, particularly large ones, are processed in an efficient manner.
In an electronic mail system, the task of processing a mailing list usually falls to a Mailing List Manager or “MLM”, such as MLM 13 for the e-mail system for
Without further enhancement to this basic process of handling an e-mail message with a large mailing list, the MLM is handing a substantial amount of work to the MTA to do, with no real intelligence. For instance, for a message sent to a predefined mailing list of 1000 recipients, the MLM is handing to the MTA a list of 1000 tasks to do in sequence—that is, 1000 messages to queue and distribute. At the same time, MTAs tend not to be very good at parallel delivery of a single message. Therefore, the approach commonly employed by MTAs is to do the tasks in series, one at a time. However, that approach incurs the penalty of increased delivery time due to network latency and/or system load.
Apart for the above one-to-many problem, an analogous problem concerns an e-mail that needs to go to a very large number of people where the e-mail's content or body is not constant but, instead, is customized for a given recipient. In such a case, one has millions of people who are intended recipients of messages that vary in content (i.e., message body)—that is, a scenario presenting a multitude of one-to-one relationships.
Present-day mass-mailing advertisers face such a problem. Doubleclick, for example, employs a “Composer” program to create customized mass e-mailings (i.e., electronic mailings). The Composer's basic operation is simple. The Composer works against a large list or database of people. Each person, in turn, has signed up to receive one or more specific topics (e.g., about travel, about business, about finance, or the like) in a regular electronic mailing or newsletter. Thus, in this large database of people, everybody has different combinations of what specific information he or she really wants. Based on this user-specific information, the Composer program will compose a customized piece of e-mail for each particular user, inserting the specific pieces of information the user has requested into the e-mail message's body and possibly even using the user's real name. After the appropriate message is composed for a given target user, the Composer directs an accompanying e-mail system to send that message to the target user. The Composer program repeats this basic operation for all individuals in its database.
For a given user, the foregoing process is relatively fast. However, a mass-mailing database may contain many millions of names. When one is faced with the task of creating customized mass e-mailings for millions of users, the approach of doing one user at a time is rather inefficient. Worse, with the standard systems that are being used today, when a system sends a message, the system waits until that message is accepted, by either the final mail server for delivery, or by an intermediate mail server enroute, before the Composer can proceed to the next address. Given the massive scale in which the operation is occurring, there is of course much interest in optimizing the process.
One approach to this problem is to attempt to run the Composer with some amount of parallelism. Here, the Composer is run in such a way that the list of recipients is broken down into smaller groups for parallel processing. For example, a group of one million people may be divided into ten groups of 100,000, each group being processed in parallel (e.g., by ten Composers running in parallel). The improvements with this approach, however, are inadequate. Reasons include excessive disk I/O (input/output) and excessive e-mail queue waiting times. The significant improvements in scalability and throughput simply are not realized with such an approach.
What is needed is an e-mail system that implements parallel processing for mass mailings, with as much resource sharing and re-use, and as little disk I/O, as possible. More particularly, it is desirable to take advantage of today's multithreaded computer systems to send e-mail on one processing thread while another processing thread is waiting for either input or a reply, including streamlining the process so that there is as little waiting as possible.
An e-mail system constructed in accordance with the present invention includes a composer module (“Composer”), a message transport agent (MTA), and a mass-mail accelerator (MMA). (The MTA component may be eliminated if all outgoing e-mail messages are processed by the MMA, as described below.) The Composer is a program that operates against a very large database of users to provide large-scale customized e-mail messages by combining different pieces of a message together on a per-user basis. Ordinarily, the Composer passes a given message on to an MTA that, in turn, transmits the message to the intended recipient. However, this basic operation is modified so that the Composer passes a given message on to the MMA, which serves to carry out e-mail delivery/routing for the messages that have been passed on to it. More particularly, the degree of parallelism on the MTA side of message delivery has been greatly increased.
In operation, the MMA receives input that, in turn, is fed into one or more queues. The input that is received, via SMTP, comprises outgoing messages from one or more Composers. A receiving (or “client”) thread initially handles this input. In the instance that multiple Composers are connected to the MMA (i.e., multiple concurrent connections), one client thread is assigned to each incoming connection. Two types of threads are actually employed here: a “listener” thread waits for a new connection, creates a client thread, and assigns the new connection to it (and thereafter repeats), and a “client” thread is what actually interacts with the Composer beyond the initial TCP/IP handshake. The respective client thread receives the incoming e-mail message (or simply, “message”) and, in turn, decides which queue from the set of queues within the MMA is appropriate to receive and process the message. Any number of queues may be supported, as desired (and as indicated by the ellipsis). The client thread that receives the message examines the configuration and state of the available queues to see which one is appropriate to receive the incoming message.
Each queue itself owns a thread that manages a list of messages. As a particular advantage, the queues themselves are configurable to either be general (generic) or be specific to a particular mail (destination) domain. For instance, a queue may be configured to handle only mail destined for the Hotmail.com domain, or configured to handle only mail destined for the AOL.com domain. A queue that is specifically configured will only handle e-mail for its specific domain and will not handle any other e-mail. In contrast, a queue may be configured to be generic or general, in which case it will handle e-mail destined for any domain which has no specific queue assigned to it. E-mail posted to a specific queue will not require a Domain Name Services (DNS) look-up, as the MMA already knows (i.e., has cached) the DNS information for the corresponding target e-mail domain. Thus, for example, e-mail destined for the AOL.com domain is posted to the AOL queue. The MMA need not look up the DNS information for the AOL.com domain as this information has already been cached as part of the setup for the AOL queue. Which queues are created is entirely dependent on the configuration which gives the customer-user (e.g., system administrator) the ability to tailor or tune for a given situation. If, for example, the system administrator knows that about 60% of outgoing e-mail for his or her company is going to AOL, then the system administrator would set up an AOL-specific queue, with corresponding resources.
Each queue manages a pool of MTA threads. During configuration of the queues, the customer's system administrator may specify the allocation of MTA threads to a given queue. For instance, a system administrator may specify a maximum and/or minimum number of MTA threads that are available to a given queue. When a given MTA thread is started, it establishes a connection out to a real MTA (e.g., remote MTA residing at a particular destination on the Internet). This connection is established using SMTP over a TCP (Transmission Control Protocol) connection. Via this connection, a given MTA thread may talk SMTP to an actual MTA out in the real world someplace (e.g., an AOL MTA).
During MMA operation, once a message has been passed to a queue, that queue examines its MTA threads to see if one is ready to accept the message. If an MTA thread is ready, the queue will assign the message to that MTA. Once a message is assigned to an MTA thread, that thread is no longer available and, thus, it marks itself as “busy” (or otherwise removes itself from a “ready” list). The MTA thread proceeds to handle the work of the SMTP exchange between the MMA and the target real-world MTA (e.g., AOL MTA). While a given MTA thread is waiting for a reply from the destination MTA (e.g., AOL MTA), the MMA can proceed to do other work. Thus, for instance, while a given message is being handled by a particular MTA thread, other incoming messages can be injected, queued, requeued, moved around, or the like, within the system. In this manner, the bottleneck usually encountered with processing mass e-mailings is removed.
The following description will focus on the presently-preferred embodiment of the present invention, which is implemented in server-based software operating in an Internet-connected environment running under a server operating system, such as the Microsoft® Windows NT running on an IBM-compatible server computer. The present invention, however, is not limited to any particular one application or any particular environment. Instead, those skilled in the art will find that the system and methods of the present invention may be advantageously embodied on a variety of different platforms, including Macintosh, Linux, BeOS, Solaris, UNIX, NextStep, FreeBSD, and the like. Therefore, the description of the exemplary embodiments that follows is for purposes of illustration and not limitation.
I. Computer-Based Implementation
A. Basic System Hardware (e.g., for Desktop and Server Computers)
The present invention may be implemented on a conventional or general-purpose computer system, such as an IBM-compatible server computer operating under a server operating system.
CPU 201 comprises a processor of the Intel Pentium® family of microprocessors. However, any other suitable microprocessor or microcomputer may be utilized for implementing the present invention. The CPU 201 communicates with other components of the system via a bi-directional system bus (including any necessary I/O controller circuitry and other “glue” logic). The bus, which includes address lines for addressing system memory, provides data transfer between and among the various components. Description of Pentium-class microprocessors and their instruction set, bus architecture, and control lines is available from Intel Corporation of Santa Clara, Calif. Random-access memory 202 serves as the working memory for the CPU 201. In a typical configuration, RAM of sixteen megabytes or more is employed. More or less memory may be used without departing from the scope of the present invention. The read-only memory (ROM) 203 contains the basic input output system code (BIOS)—a set of low-level routines in the ROM that application programs and the operating systems can use to interact with the hardware, including reading characters from the keyboard, outputting characters to printers, and so forth.
Mass storage devices 215, 216 provide persistent storage on fixed and removable media, such as magnetic, optical or magnetic-optical storage systems, or flash memory, or any other available mass storage technology. The mass storage may be shared on a network, or it may be a dedicated mass storage. As shown in
In basic operation, program logic (including that which implements methodology of the present invention described below) is loaded from the storage device or mass storage 216 into the main (RAM) memory 202, for execution by the CPU 201. During operation of the program logic, the system 200 accepts user input from a keyboard 206 and pointing device 208, as well as speech-based input from a voice recognition system (not shown). The keyboard 206 permits selection of application programs, entry of keyboard-based input or data, and selection and manipulation of individual data objects displayed on the display screen 205. Likewise, the pointing device 208, such as a mouse, track ball, pen device, or the like, permits selection and manipulation of objects on the display screen. In this manner, these input devices support manual user input for any process running on the system.
The computer system displays text and/or graphic images and other data on the display device 205. The video adapter 204, which is interposed between the display 205 and the system, drives the display device 205. The video adapter 204, which includes video memory accessible to the CPU, provides circuitry that converts pixel data stored in the video memory to a raster signal suitable for use by a cathode ray tube (CRT) raster or liquid crystal display (LCD) monitor. A hard copy of the displayed information, or other information within the system 200, may be obtained from the printer 207, or other output device. Printer 207 may include, for instance, an HP Laserjet® printer (available from Hewlett-Packard of Palo Alto, Calif.), for creating hard copy images of output of the system.
The system itself communicates with other devices (e.g., other computers) via the network interface card (NIC) 211 connected to a network (e.g., Ethernet network), and/or modem 212 (e.g., 56K baud, ISDN, DSL, or cable modem), examples of which are available from 3Com of Santa Clara, Calif. The system 200 may also communicate with local occasionally-connected devices (e.g., serial cable-linked devices) via the communication (“comm”) interface 210, which may include a RS-232 serial port, a Universal Serial Bus (USB) interface, or the like. Other devices may be connected locally via the comm interface.
IBM-compatible personal computers and server computers are available from a variety of vendors. Representative vendors include Dell Computers of Round Rock, Tex., Compaq Computers of Houston, Tex., and IBM of Armonk, N.Y. Other suitable computers include Sun Solaris workstations, which are available from Sun Microsystems of Mountain View, Calif.
The above-described system 200 of
B. Basic System Software
System 300 includes a graphical user interface (GUI) 315, for receiving user commands and data in a graphical (e.g., “point-and-click”) fashion. These inputs, in turn, may be acted upon by the system 200 in accordance with instructions from operating system 310, and/or client application module(s) 301. The GUI 315 also serves to display the results of operation from the OS 310 and application(s) 301, whereupon the user may supply additional inputs or terminate the session. Typically, the OS 310 operates in conjunction with device drivers 320 (e.g., “Winsock” driver) and the system BIOS microcode 330 (i.e., ROM-based microcode), particularly when interfacing with peripheral devices. OS 310 can be provided by a conventional operating system, such as Microsoft® Windows 9x, by Microsoft® Windows NT, or by Microsoft® Windows 2000, all available from Microsoft Corporation of Redmond, Wash. Alternatively, OS 310 can also be an alterative operating system, such as IBM OS/2 (available from IBM of Armonk, N.Y.) or Linux OS (available from several vendors, including the Red Hat distribution of Linux from Red Hat, Inc. of Durham, N.C.).
The above-described computer hardware and software are presented for purposes of illustrating the basic underlying desktop and server computer components that may be employed for implementing the present invention. For purposes of discussion, the following description will present examples in which it will be assumed that there exists an MTA (e.g., sendmail) that communicates with one or more other MTAs (e.g., remote MTAs), as well as other relevant systems (e.g., Composer, which is described below). The present invention, however, is not limited to any particular environment or hardware configuration. In particular, an MTA distinction is not necessary to the invention, but is used to provide a framework for discussion. Instead, the present invention may be implemented in any type of system architecture or processing environment capable of supporting the methodologies of the present invention presented in detail below.
II. Accelerating Mass Mailings
A. Overview of System Components and Basic Operation
The lower half of
In order to understand the approach adopted by the MMA 430, it is instructive to first consider the basic transactions that occur in an e-mail system. As shown in
In accordance with the present invention, basic operation of the e-mail system is modified as follows. As shown by the transaction labeled “C” in the figure, the transaction for a given e-mail message may instead comprise transferring the message from the Composer 410 to the MMA 430. Now, in contrast to the above, the operation or transaction of recipient delivery, labeled as “D”, begins while C is allowed to complete immediately. Thus, an additional instance of C can start while D is in progress. Continuing in this manner, multiple instances of Transaction C may commence regardless of whether corresponding instances of Transaction D have concluded; this approach introduces a pronounced degree of parallelism into the system 400. Further, as described below, the approach may be fine-tuned to further accentuate the degree of parallelism and overall performance that the system may achieve. Notwithstanding these modifications, the system still employs the SMTP protocol between system components. Although the internal operations between Transactions C and D differ substantially from those of Transactions A and B, the internal operations are performed in a manner that is invisible to the SMTP protocol, thereby allowing the improved system to maintain compatibility and simplicity afforded by the SMTP protocol.
B. Connection Caching
In order to improve throughput, the system of the present invention extends the established “connection caching” technique, initially employed in the MTA, to include connections made by the MMA as well. Consider, for instance, a message destined for an AOL user (i.e., a user in the AOL.com domain). The MTA, without the enhancements of the present invention, would simply make a connection to AOL's mail server to deliver that message, and thereafter terminate the connection. However, the act of looking up AOL's mail server (e.g., in a domain name server) and making a connection (e.g., establishing the TCP/IP connection and exchanging the initial protocol commands) is resource-expensive and time-consuming. At the same time, in the process of going through its queue and seeing how much work it has to do, the MTA might discover that it has more than one message destined for the AOL domain. Therefore, instead of terminating the just-made connection to AOL's mail server in the above example, the system of the present invention would keep the connection open for use in a connection cache. More particularly, the connection cache keeps connections open to a configurable number of the most-recently connected-to domains. In the currently-preferred embodiment, the configurable number defaults to five (5); that number may be modified, as desired. In response to this setting, the system keeps the five most-recent connections open, with connections being “aged” or rotated out using a least-recently used (LRU) technique (i.e., the least-recently used items are removed to make room for new entries). Thus, in the foregoing example, if the system encounters a number of messages sent to the AOL domain, the connection cache serves to keep the connection to AOL open, thereby allowing the system to reuse that connection without the expense normally associated with obtaining a connection for a given message.
In a similar manner, the technique can be applied to Domain Name Service (DNS) resolution. Domain Name Service is an Internet service that translates domain names into IP addresses. Domain names are alphabetic, so that they are easier to remember. The Internet, however, is really based on numeric IP addresses. Therefore, every time one employs a domain name, a DNS service must translate the name into the corresponding IP address. In the context of an e-mail system, the e-mail system must query a DNS server to determine the e-mail server for a given domain (e.g., aol.com or a hotmail.com). Again, the process is time-consuming, as the e-mail system must wait for the DNS server to respond. By storing this information in a domain name cache, the e-mail system may avoid repeating the query for already-known domains and thus avoid the delay typically incurred while waiting for the information to be returned from a DNS server.
C. MMA Detailed Architecture
The input 501 receives, via SMTP, outgoing messages from one or more Composers (previously described). A client thread 503 initially handles this input. In the instance that multiple Composers are connected to the MMA 500 (i.e., multiple concurrent connections), one client thread is assigned to each incoming connection. The respective client thread receives the incoming e-mail message (or simply, “message”) and, in turn, decides which queue from the set of queues 510 within the MMA 500 is appropriate to receive and process the message. Although the figure illustrates only two queues, any number of queues may be supported, as desired (and as indicated by the ellipsis). The client thread that receives the message examines the available queues (e.g., queue 511, queue 513) to see which one is appropriate to receive the incoming message.
Each queue itself owns a thread that manages a list of messages. As a particular advantage, the queues themselves are configurable to either be general (generic) or be specific to a particular mail (destination) domain. For instance, a queue may be configured to handle only mail destined for the Hotmail.com domain, or configured to handle only mail destined for the AOL.com domain (e.g., as shown by queue 513). A queue that is specifically configured will only handle e-mail for its specific domain and will not handle any other e-mail. In contrast, a queue may be configured to be generic or general, in which case it will handle e-mail destined for any domain which has no specific queue assigned to it. E-mail posted to a specific queue will not require a Domain Name Services (DNS) look-up, as the MMA 500 already knows (i.e., has cached) the DNS information for the corresponding target e-mail domain. Thus, for example, e-mail destined for the AOL.com domain is posted to the AOL queue 513. The MMA 500 need not look up the DNS information for the AOL.com domain as this information has already been cached as part of the setup for the AOL queue 513. Which queues are created is entirely dependent on the configuration which gives the customer-user (e.g., system administrator) the ability to tailor or tune for a given situation. If, for example, the system administrator knows that about 60% of outgoing e-mail for his or her company is going to AOL, then the system administrator would set up an AOL-specific queue, with corresponding resources.
Each queue manages a pool of MTA threads. For example as illustrated in the figure, the general or generic queue 511 manages MTA threads 521; the AOL queue 513 manages the MTA threads 523. During configuration of the queues, the customer's system administrator may specify the allocation of MTA threads to a given queue. For instance, a system administrator may specify a maximum and/or minimum number of MTA threads that are available to a given queue. When a given MTA thread is started, it establishes a connection out to a real MTA (e.g., remote MTA residing at a particular destination on the Internet). This connection is established using SMTP over a TCP (Transmission Control Protocol) connection, such as TCP connection 530. Via this connection, a given MTA thread may talk SMTP to an actual MTA out in the real world someplace (e.g., AOL MTA).
During MMA operation, once a message has been passed to a queue, that queue examines its MTA threads to see if one is ready to accept the message. If an MTA thread is ready, the queue will assign the message to that MTA thread (which exists inside the MMA 500). Once a message is assigned to an MTA thread, that thread is no longer available and, thus, it marks itself as “busy” (or otherwise removes itself from a “ready” list). The MTA thread proceeds to handle the work of the SMTP exchange between the MMA and the target real-world MTA (e.g., AOL MTA). While a given MTA thread is waiting for a reply from the destination MTA (e.g., AOL MTA), the MMA can proceed to do other work. Thus, for instance, while a given message is being handled by a particular MTA thread, other incoming messages can be injected, queued, requeued, moved around, or the like, within the system.
If, when processing an incoming message, a given queue thread finds all of its MTA threads busy, the queue thread may launch another MTA thread (unless the queue thread has reached a user-specified maximum number of corresponding MTA threads). The newly created thread will then proceed to connect to the destination MTA and attempt delivery. This process of spawning new MTA threads may continue until underlying resources of the base hardware system are exhausted (e.g., system has run out of file descriptors or memory). In instances where no more MTA threads can be created, the system logs corresponding information to a log file, thereby allowing the system administrator to fine-tune the underlying system (e.g., adjust the balance of queue and MTA resources) for the next run.
In contrast to a specific queue, the general or “others” queue is not specific to a particular domain. Thus, the general queue receives messages that are not posted to one of the specific queues. The general queue employs, in addition to a pool of MTAs, the above-mentioned connection-caching technique, including maintaining a list of name server data (i.e., DNS data), as illustrated in
Since the general queue talks to domains that are not serviced by a domain-specific queue, its treatment of connections is slightly different because it might need to throw away an old connection in favor of creating a new one to something it has either never talked to, or has not talked to recently. Consider, for instance, the scenario where the general queue is employing a user-specified maximum number of MTAs, which are currently bound to one or more destination MTAs, when a new message arrives that is destined for the domain xyz.com for which the general queue does not have any connection open. In this case, the general queue will shut down an old connection so that it can make a new one (i.e., within the user-specified maximum number of MTAs). Here, the general queue will query a DNS server for the appropriate name server data for the newly-encountered domain. Upon receiving the name server data, the general queue invokes an MTA thread for connecting to the newly-encountered domain, whereupon the MTA thread may transmit the message destined for the xyz.com domain to the real-world MTA for xyz.com. If the general queue encounters a particular domain on a repeated basis, the general queue may post a corresponding entry to the MMA's log file to alert the user that it may be appropriate to establish a specific queue for this particular domain. This information may also be uncovered by performing a database query against the database of recipients, with the results being sorted by domain name.
A queue can be configured to pass messages it cannot complete to some other queue for special handling. This might be caused, for example, by a total network failure between the MMA and a particular destination, such as AOL.com or hotmail.com, as in previous examples. Where this is the case, the queue thread which decides it is unable to complete the delivery will hand the message to its designated “retry” queue, and that queue will attempt to deliver the message using the MTAs and other features available to it. If the initial queue thread has no “retry” queue assigned, the message fails completely, an error is logged, and the message is discarded (or in safe mode, the SMTP submission into the MMA fails).
A particular case of a useful “retry” queue is one which is configured to connect to local MTAs which do nothing but queue messages to disk for later attempts to connect. Such an MTA is said to be operating in “deferred” or “fallback” mode.
D. Detailed Internal Operation
1. Data Structures
Before describing internal methods of operation of the present intention in detail, it is first instructive to review data structures employed within the system. In particular, a “queue” handle is defined to be an instance of the following data structure.
2: ** QUEUE handle, defining a queue of work for assignment to MTAs
5: struct queue_handle
7: unsigned q_randctx; /* rand_r( ) context */
8: u_int q_mtaid; /* MTA sequence number */
9: u_int q_initmta; /* initial MTAs to be created */
10: u_int q_maxmta; /* max. MTAs to be created */
11: u_int q_maxfail; /* max. MTA failures */
12: u_int q_maxjobs; /* max. MTA jobs */
13: u_int q_maxqueue; /* max. queued messages before
14: creating a new MTA */
15: u_int q_flags; /* flags */
16: u_int q_startwait;/* secs to wait for MTA start */
17: u_int q_peakmtas; /* peak MTA count */
18: u_int q_peakmsgs; /* peak MSG count */
19: u_int q_mtatimeout;/* MTA timeout (secs) */
20: u_int q_batchsize;/* max. rcpts per transaction */
21: u_int q_mxlimit; /* MX record count limit */
22: #ifdef _FFR_THROTTLE
23: u_int q_throttle; /* max. queue length */
24: #endif /* _FFR_THROTTLE */
25: #ifdef _FFR_NEW_TIMEOUTS
26: u_int q_mtadatatimeout; /* MTA DATA timeout (secs) */
27: u_int q_maxconntime; /* max. time to connect */
28: u_int q_maxtranstime; /* max. SMTP session time */
29: u_int q_maxdelivertime; /* max. time to deliver */
30: #endif /* _FFR_NEW_TIMEOUTS */
31: #ifdef _FFR_MX_HANDLE
32: u_int q_mxtimeout;/* max. time to keep MX list */
33: #endif /* _FFR_MX_HANDLE */
34: u_long q_msgcount; /* messages processed */
35: #ifdef _FFR_MAX_MESSAGE_SIZE
36: u_long q_maxsize; /* max. message size */
37: #endif /* _FFR_MAX_MESSAGE_SIZE */
38: char * q_name; /* name of the queue */
39: Arena q_arena; /* arena for memory stuff */
40: QUEUE q_retry; /* pointer to retry queue */
41: #ifdef _FFR_DMS_MANAGER
42: DNSMGR q_dnsmgr; /* DNS manager */
43: #endif /* _FFR_DNS_MANAGER */
44: Vector q_mtas; /* connected MTAs */
45: Vector q_ready; /* MTAs ready for work */
46: Vector q_msgs; /* queued messages */
47: Vector q_hosts; /* hosts for connect( ) */
48: HashTable q_hosthash; /* hosts for connect( ) */
49: #ifdef _FFR_TEST_MODE
50: HashTable q_testmtas;/* test MTAs */
51: #endif /* _FFR_TEST_MODE */
52: #ifdef _FFR_MX_HASH
53: HashTable q_mxhash; /* MX list */
54: #endif /* _FFR_MX_HASH */
55: #ifdef _FFR_DNS_AVOID_DUP
56: HashTable q_dnsip; /* DNS queries in progress */
57: #endif /* _FFR_DNS_AVOID_DUP */
58: pthread_t q_thread; /* thread running this queue */
59: pthread_cond_t q_gotwork; /* work to do */
60: pthread_cond_t q_gotmta; /* MTA now ready */
61: pthread_mutex_t q_lock; /* lock for this queue */
As shown, the data structure fully characterizes a given queue (instance). For example, the q_name member specifies a text string for the queue's name. The q_maxsize member specifies the maximum message size handled by the queue instance, and the q_msgcount member records a count of messages handled by the queue instance. The q_thread member indicates the thread running the particular queue instance. Several members specify MTA characteristics associated with a given queue instance. For example, the q_initmta member specifies an initial number of MTA threads to be created for a given queue instance. The q_maxqueue member, on the other hand, indicates the maximum number of messages waiting before a new MTA thread is created to process them. The basic function of other members of the data structure may be discerned from the included programming comments associated with each member.
In a similar manner, data structures may be defined for characterizing a message (object) and MTA (instance), as follows.
2: ** MSG handle, defining a message which has not been processed
5: struct msg_handle
7: bool * msg_success;/* creator's success flag */
8: int msg_flags; /* flags */
9: #ifdef _FFR_DNS_MANAGER
10: u_int msg_dnsretry;/* number of DNS retries */
11: #endif /* _FFR_DNS_MANAGER */
12: u_int msg_retries;/* retry count */
13: #ifdef _FFR_DISK_CACHE
14: u_long msg_memsize;/* max. in-memory body */
15: #endif /* _FFR_DISK_CACHE */
16: Arena msg_arena; /* memory arena */
17: char * msg_jobid; /* job ID for logging */
18: char * msg_client; /* client, from LHLO */
19: SENDER msg_sender; /* sender */
20: Vector msg_rcpt; /* RECIPIENT vector */
21: #ifdef _FFR_MTA_HISTORY
22: Vector msg_mtahistory; /* MTA history */
23: #endif /* _FFR_MTA_HISTORY */
24: BODY msg_body; /* body */
25: QUEUE msg_queue; /* queue handling this msg */
26: MTA msg_mta; /* MTA handling this msg */
27: pthread_cond_t * msg_donesig; /* done signal */
28: struct timeval msg_queuetime; /* queue time */
29: struct timeval msg_start; /* start time */
30: #ifdef _FFR_DNS_MANAGER
31: struct timeval msg_dnsqueue; /* queue time */
32: struct timeval msg_dnsstart; /* DNS query start time */
33: struct timeval msg_dnsdone; /* DNS query done time */
34: #endif /* _FFR_DNS_MANAGER */
35: #ifdef _FFR_NEW_TIMEOUTS
36: struct timeval msg_aborttime; /* time after which to punt */
37: #endif /* _FER_NEW_TIMEOUTS */
38: struct timeval msg_done; /* finish time */
42: ** MTA handle, defining a remote MTA
45: struct mta_handle
47: char mta_name[MAXMTANAME + 1];
/* connected host or prog */
48: u_int mta_flags; /* status */
49: int mta_crlf; /* CRLF handling */
50: int mta_infd; /* input fd */
51: int mta_outfd; /* output fd */
52: pid_t mta_pid; /* pid of child */
53: int mta_exstatus; /* child exit status */
54: u_long mta_msgcount; /* messages handled */
55: u_int mta_timeout; /* input wait timeout */
56: #ifdef _FFR_NEW_TIMEOUTS
57: u_int mta_datatimeout; /* input wait timeout */
58: #endif /* _FFR_NEW_TIMEOUTS */
59: u_int mta_failed; /* failed message count */
60: HOST mta_host; /* HOST handle */
61: QUEUE mta_queue; /* queue to which I belong */
62: MSG mta_msg; /* message I'm working on */
63: #ifdef _FFR_TEST_MODE
64: TEST mta_test; /* TEST handle */
65: #endif /* _FFR_TEST_MODE */
66: #ifdef _FFR_LISTENER
67: CLIENT mta_client; /* CLIENT handle */
68: #endif /* _FFR_LISTENER */
69: Vector mta_esmtp; /* ESMTP services available */
70: struct in_addr mta_addr; /* IP address in use */
71: struct timeval mta_conntime; /* when connected */
72: struct timeval mta_lastused; /* when last used */
73: #ifdef _FFR_NEW_TIMEOUTS
74: struct timeval mta_aborttime; /* time after which to abort */
75: #endif /* _FFR_NEW_TIMEOUTS */
76: pthread_t mta_thread; /* thread running this MTA */
77: pthread_cond_t mta_gotwork; /* condition to start work */
78: pthread_mutex_t mta_lock; /* structure lock */
As shown, a message (MSG) handle data structure is defined at lines 1–39; it specifies the data members characterizing an incoming message that is to be processed. An MTA handle data structure is defined at lines 41–79, for characterizing a remote or real-world (destination) MTA.
2. Methods of Operation
a) Core Operation
The method begins with the first invocation of the MMA, for instance, from a Composer program that has already started. Here, the Composer will make a connection to the MMA, as shown at step 601. This is done through a UNIX-style pipe, or through a socket (e.g., TCP/IP), or through any other construct/process that allows data to pass back and forth (i.e., supporting interprocess communication). Now, the Composer will begin delivering a message by initiating an SMTP dialog/session with the MMA, as shown in step 602. At step 603, SMTP phase processing begins; this phase is repeated for each message. Here, a given e-mail message is parsed (e.g., for determining its sender, recipient, body, attachments, delivery restrictions, and the like). After the message is parsed, it is ready for delivery. At the completion of the SMTP phase for the first recipient, the MMA now has the message proper (e.g., accessible via a handle) and all of its corresponding parameters required for delivery, and may therefore return an acknowledgment back to the Composer that the message has been successfully received and that the MMA is now ready for more messages.
Operation at this point is controlled based on how the MMA is configured: (1) safe mode or (2) unsafe mode. In safe mode, the MMA will not send an acknowledgement back until the message has passed through the MMA in its entirety—that is, that the message was successfully sent somewhere, such that some other MTA in the world received and accepted responsibility for the message. In other words, in safe mode, only when the MMA successfully sends the message (to another MTA) will it indicate to the Composer that it is ready for more work (i.e., to receive additional messages). Safe mode is provided to comply strictly with the definitions of RFC 821 (SMTP), which guarantees message reliability. However, this atomic transaction-based approach presents a bottleneck to system throughput. As a result, the MMA is also allowed to operate in unsafe mode.
In unsafe mode, the MMA will indicate to the Composer that it is ready for more work regardless of whether the current message has already been successfully sent somewhere. As a result, unsafe mode operates a great deal faster because there is no waiting for confirmation that a given message has been successfully received by another MTA. Although unsafe mode may provide less reliability (e.g., no guarantee of service), the mode operates a great deal faster since the bottleneck of awaiting confirmation is removed, thereby allowing a high degree of parallelism to be realized. The system is user-configurable so that each individual customer can decide which mode is preferred. Given the substantial benefits in performance, however, most users will typically use the unsafe mode of operation. Thus in typical operation, where the system is operating in unsafe mode, step 603 is allowed to loop, or repeatedly execute, for a multitude of messages.
As shown at step 604, each incoming message is received by a client thread, which is listening for incoming messages from the Composer. (The client thread is distinguishable from the “listener” thread, which exists in the MMA only to listen for new connections.) Upon receiving a new incoming message, the client thread decides to which queue or queues it will assign the workload of processing the message. In a typical case, a given message will have only a single recipient. For example, the message may be addressed to an AOL recipient. In that case, the client thread determines whether there are any queues that specifically process AOL messages (i.e., messages destined for the AOL.com domain). If an available specific queue is found, the client thread assigns the message to that queue. Otherwise, the message will be assigned to the general queue, as shown at step 604.
A less common case is a message with multiple recipients. Here, the client thread in effect “clones” the message and assigns the clones to the appropriate queues. For example, if a given message is addressed to an AOL recipient and a xyz recipient, the workload for the AOL recipient is assigned to the AOL queue and the xyz recipient is assigned to the general queue. During this “cloning” process, the message body itself is not needlessly duplicated. Instead, only a single copy of the message body exists. Each clone refers to the message body via a reference handle.
As shown at step 605, the act of assigning work to a queue thread awakens that thread. Upon awakening, a queue thread may now proceed with its assigned work. Specifically, the awakened queue thread looks to see if it has any MTA threads that are ready to handle the job that has been assigned, as shown at step 606. Here, there is a distinction between MTA threads that are running and MTA threads that are ready. In the currently-preferred embodiment, each MTA thread has an “inbox” for receiving a single job at a time. Once a job is placed in the MTA thread's inbox, that MTA thread is removed from the queue thread's list of ready MTA threads, as shown at step 607 in
Having received a new job, the MTA thread, in turn, communicates with a real-world MTA (to which it is connected), using the SMTP protocol, in order to execute the delivery, as shown at step 609. Here, the same set of SMTP commands that the Composer used to post a message to the MMA are, in turn, sent directly to the destination. Now, at step 610, the particular remote (real-world) MTA takes over responsibility for delivery of the message. Once the remote MTA confirms that it will take over responsibility for delivery, the MTA thread (which has handled this message) can now return itself to the “ready” list, as shown at step 611.
b) Exception Case #1: No MTA Threads are Available
Suppose in step 606 that no MTA threads are available. In that case, processing proceeds as shown in
c) Exception Case #2: Error Condition Exists at Remote (Real-World) MTA
d) Exception Case #3: MTA Not Available for Domain
Recall that the general queue handles all domains for which there is not a specific queue. Thus, the general queue must make sure that there is a real-world MTA available for a particular new domain that is encountered. Consider, for instance, a scenario in which the general queue receives a message destined for the xyz.com domain but it has not opened a connection to the real-world MTA yet. In that instance, the general queue must open such a connection. The processing proceeds as outlined in
At step 901, the system attempts to locate an MTA thread handling mail for the recipient domain. If one is found, the workload for delivering the message is assigned to that MTA thread (whereupon the method may return). However, if one is not found, the method proceeds to step 902. The method locates the MTA (host) that handles e-mail for this domain. That information may already exist in the DNS cache. Therefore, at step 902, the method checks the DNS cache for an entry corresponding to the domain. If that information is not already stored in the DNS cache, the MMA system must look up that information on the Internet, by querying a DNS server, as indicated by step 903. Two queries are actually performed: a first query for determining which machine handles mail for that domain and a second query for determining that machine's IP address. Once the DNS information is obtained (either from its cache or from querying a DNS server), a new MTA thread makes a connection to the real-world MTA that handles mail for the recipient domain, at step 904, including performing the initial SMTP handshake. Now, the work can be assigned to the new MTA thread, as indicated at step 905. After message delivery, that MTA thread will remain in the general queue's pool of MTA threads until it is instructed to shut down and make room for something else (e.g., an MTA thread for another domain).
While the invention is described in some detail with specific reference to a single-preferred embodiment and certain alternatives, there is no intent to limit the invention to that particular embodiment or those specific alternatives. For instance, although the currently-preferred embodiment has been described in terms of receiving input originating from a Composer program, the system of the present invention may receive input from any program capable of generating mass e-mailings. There is no requirement that a Composer program be used. All told, those skilled in the art will appreciate that modifications may be made to the preferred embodiment without departing from the teachings of the present invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5463620 *||Feb 22, 1994||Oct 31, 1995||At&T Ipm Corp.||Bandwidth allocation, transmission scheduling, and congestion avoidance in broadband asynchronous transfer mode networks|
|US5937162 *||Sep 24, 1996||Aug 10, 1999||Exactis.Com, Inc.||Method and apparatus for high volume e-mail delivery|
|US6148329 *||Jul 20, 1998||Nov 14, 2000||Unisys Corporation||Method and system for maintaining the format of messages in a messaging system database|
|US6502131 *||Dec 4, 1998||Dec 31, 2002||Novell, Inc.||Directory enabled policy management tool for intelligent traffic management|
|US6658454 *||Feb 7, 2000||Dec 2, 2003||Sendmail, Inc.||Electronic mail system with improved methodology for processing messages with mailing lists|
|US6735770 *||Apr 27, 1998||May 11, 2004||Sun Microsystems, Inc.||Method and apparatus for high performance access to data in a message store|
|US6779039 *||Mar 31, 2000||Aug 17, 2004||Avaya Technology Corp.||System and method for routing message traffic using a cluster of routers sharing a single logical IP address distinct from unique IP addresses of the routers|
|US20020129127 *||Mar 6, 2001||Sep 12, 2002||Romero Francisco J.||Apparatus and method for routing a transaction to a partitioned server|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7502937||Mar 4, 2003||Mar 10, 2009||Digimarc Corporation||Digital watermarking security systems|
|US7590700 *||Aug 30, 2005||Sep 15, 2009||Ntt Docomo, Inc.||Email multicasting device|
|US7698369 *||May 27, 2004||Apr 13, 2010||Strongmail Systems, Inc.||Email delivery system using metadata on emails to manage virtual storage|
|US7716217||Jan 16, 2007||May 11, 2010||Bluespace Software Corporation||Determining relevance of electronic content|
|US7958359||Apr 30, 2001||Jun 7, 2011||Digimarc Corporation||Access control systems|
|US8073911||May 12, 2005||Dec 6, 2011||Bluespace Software Corporation||Enforcing compliance policies in a messaging system|
|US8108476 *||Jun 1, 2009||Jan 31, 2012||Quest Software, Inc.||High performance electronic message delivery engine|
|US8352558||Feb 10, 2009||Jan 8, 2013||Microsoft Corporation||Transport high availability via acknowledge management|
|US8364769 *||Mar 22, 2010||Jan 29, 2013||Teamon Systems, Inc.||Communications system providing message aggregation features and related methods|
|US8402100||Apr 12, 2010||Mar 19, 2013||Strongmail Systems, Inc.||Email delivery system using metadata on emails to manage virtual storage|
|US8584211||May 18, 2011||Nov 12, 2013||Bluespace Software Corporation||Server-based architecture for securely providing multi-domain applications|
|US8805944||Dec 10, 2012||Aug 12, 2014||Microsoft Corporation||Transport high availability via acknowledge management|
|US8914455||Mar 18, 2013||Dec 16, 2014||Strongview Systems, Inc.||Systems and methods for processing emails|
|US9021559||Oct 11, 2013||Apr 28, 2015||Bluespace Software Corporation||Server-based architecture for securely providing multi-domain applications|
|US9258259 *||Feb 9, 2006||Feb 9, 2016||Nokia Technologies Oy||Retrieval of offline instant messages|
|US20040128512 *||Apr 30, 2001||Jul 1, 2004||Sharma Ravi K||Digital watermarking systems|
|US20040243806 *||Mar 4, 2003||Dec 2, 2004||Mckinley Tyler J.||Digital watermarking security systems|
|US20050144242 *||Oct 28, 2004||Jun 30, 2005||Justin Marston||Caching in an electronic messaging system|
|US20050267941 *||May 27, 2004||Dec 1, 2005||Frank Addante||Email delivery system using metadata on emails to manage virtual storage|
|US20060031351 *||May 12, 2005||Feb 9, 2006||Justin Marston||Enforcing compliance policies in a messaging system|
|US20060031352 *||May 12, 2005||Feb 9, 2006||Justin Marston||Tamper-proof electronic messaging|
|US20060053206 *||Aug 30, 2005||Mar 9, 2006||Masaki Hirose||Email multicasting device|
|US20070078935 *||Feb 9, 2006||Apr 5, 2007||Nokia Corporation||Retrieval of offline instant messages|
|US20070179945 *||Jan 16, 2007||Aug 2, 2007||Bluespace Software Corporation||Determining relevance of electronic content|
|US20090157817 *||Dec 12, 2007||Jun 18, 2009||International Business Machines Corporation||Using an unsynchronized event pool to improve performance of an event driven im gateway|
|US20090240780 *||Jun 1, 2009||Sep 24, 2009||Brown Scott T||High Performance Electronic Message Delivery Engine|
|US20100009714 *||Mar 10, 2009||Jan 14, 2010||Mckinley Tyler J||Decoding Information to Allow Access to Computerized Systems|
|US20100179999 *||Mar 22, 2010||Jul 15, 2010||Teamon Systems, Inc.||Communications system providing message aggregation features and related methods|
|US20100205257 *||Feb 10, 2009||Aug 12, 2010||Microsoft Corporation||Transport high availability via acknowledge management|
|US20100306321 *||May 29, 2009||Dec 2, 2010||Microsoft Corporation||Delivering messages using user-defined agents|
|U.S. Classification||709/206, 718/107|
|International Classification||G06F9/46, H04L12/58, H04L12/18, G06F15/16|
|Cooperative Classification||H04L51/14, H04L12/1863, H04L12/5855|
|European Classification||H04L12/58G, H04L51/14|
|May 22, 2001||AS||Assignment|
Owner name: SENDMAIL, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:KUCHERAWY, MURRAY;REEL/FRAME:011848/0359
Effective date: 20010521
|Oct 9, 2007||AS||Assignment|
Owner name: PARTNERS FOR GROWTH II, L.P., CALIFORNIA
Free format text: SECURITY AGREEMENT;ASSIGNOR:SENDMAIL, INC.;REEL/FRAME:019930/0306
Effective date: 20071009
|Nov 14, 2007||AS||Assignment|
Owner name: PARTNERS FOR GROWTH II, L.P., CALIFORNIA
Free format text: SECURITY AGREEMENT;ASSIGNOR:SENDMAIL, INC.;REEL/FRAME:020105/0620
Effective date: 20071112
|Jan 7, 2008||AS||Assignment|
Owner name: SILICON VALLEY BANK, CALIFORNIA
Free format text: SECURITY AGREEMENT;ASSIGNOR:SENDMAIL, INC.;REEL/FRAME:020317/0891
Effective date: 20071219
|Jun 12, 2009||FPAY||Fee payment|
Year of fee payment: 4
|Feb 4, 2013||AS||Assignment|
Owner name: GOLD HILL CAPITAL 2008, LP, CALIFORNIA
Free format text: SECURITY AGREEMENT;ASSIGNOR:SENDMAIL, INC.;REEL/FRAME:029745/0107
Effective date: 20130131
|Jul 16, 2013||FPAY||Fee payment|
Year of fee payment: 8
|Jul 26, 2013||AS||Assignment|
Effective date: 20130726
Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:PARTNERS FOR GROWTH II, L.P.;REEL/FRAME:030884/0540
Owner name: SENDMAIL, INC., CALIFORNIA
|Oct 17, 2013||SULP||Surcharge for late payment|
|Nov 15, 2013||AS||Assignment|
Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:SILICON VALLEY BANK;REEL/FRAME:031616/0661
Owner name: SENDMAIL, INC., CALIFORNIA
Effective date: 20131001
|Nov 26, 2013||AS||Assignment|
Owner name: SENDMAIL, INC., CALIFORNIA
Effective date: 20131001
Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:DECATHLON ALPHA, L.P.;REEL/FRAME:031679/0972
Owner name: PROOFPOINT, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:SENDMAIL, INC.;REEL/FRAME:031682/0572
Effective date: 20131122
Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:GOLD HILL CAPITAL 2008, LP;REEL/FRAME:031679/0643
Effective date: 20131001
Owner name: SENDMAIL, INC., CALIFORNIA